Online fault diagnosis method and device for power electronic converter and electronic equipment

Abstract

The invention provides an online fault diagnosis method and device for a power electronic converter and electronic equipment, and the method comprises the steps: obtaining a data sequence of three-phase output current outputted by a target converter; performing multi-step moving average processing on the data sequence to obtain a feature matrix; calling an online classifier, performing fault diagnosis on the target converter based on the feature matrix, and generating a fault diagnosis result; the online classifier is obtained by training a pre-constructed initial classifier through dual-core extreme learning machine hyper-parameter common factor optimization. According to the invention, the data size is reduced, and the processing difficulty of the online classifier can be reduced, so that the fault diagnosis speed of the target converter is improved, and the problem of low fault diagnosis efficiency in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power electronics, and in particular to an online fault diagnosis method, device and electronic equipment for a power electronic converter. Background Art

[0002] Multilevel converters have better application potential in medium and low voltage or even low voltage high power scenarios, such as motor drive and large wind power generation, due to lower switching losses and smaller output voltage harmonics. Among multilevel converters, the four-level active neutral point clamped (4L-ANPC) converter topology has the characteristics of fewer switching devices and simple circuit configuration, and is considered to be the "simplest" four-level converter topology. Therefore, the 4L-ANPC converter has better application prospects. However, power semiconductor devices are prone to failure, especially short circuit and open circuit fault (OCF). Among them, OCF does not affect the normal operation of the converter for a period of time, but over time, it may cause damage to the converter load and even cause the DC link capacitor to break down. Therefore, it is of great significance to carry out research on OCF diagnosis of switching devices of 4L-ANPC converters.

[0003] In recent years, data-driven converter diagnosis methods have become a research hotspot. However, model training of this method is very time-consuming, and the diagnostic accuracy is highly dependent on data quality, making it difficult to implement online applications. In addition, highly noisy measured data increases the training of high-precision model diagnosis. In order to solve the problem that the accuracy of data-driven models depends on data quality, existing methods deal with this problem through statistical learning and signal fusion methods. Although the influence of data noise has been overcome to a certain extent, the time-consuming training of data-driven models still makes it difficult to implement online diagnosis.

[0004] There is no effective solution to the problem of low fault diagnosis efficiency in the existing related technologies. Summary of the invention

[0005] The present invention provides a method, a device and an electronic device for online fault diagnosis of a power electronic converter, which are used to solve the defect of low fault diagnosis efficiency in the prior art.

[0006] In a first aspect, the present invention provides a method for online fault diagnosis of a power electronic converter, comprising: Acquire a data sequence of three-phase output current output by the target converter; Performing multi-step sliding average processing on the data sequence to obtain a feature matrix; An online classifier is called to perform fault diagnosis on the target converter based on the characteristic matrix to generate a fault diagnosis result; the online classifier is obtained by training a pre-built initial classifier through a dual-core extreme learning machine hyperparameter common factor optimization.

[0007] According to an online fault diagnosis method for a power electronic converter provided by the present invention, a multi-step sliding average process is performed on the data sequence to obtain a characteristic matrix, including: Performing unit step sliding average processing on the data sequence to remove noise points in the data sequence; The data sequence is subjected to large-step sliding equidistant averaging processing to obtain a characteristic matrix of the data sequence.

[0008] According to a method for online fault diagnosis of a power electronic converter provided by the present invention, unit step sliding average processing is performed on the data sequence, comprising: A sliding window is set, which slides with a unit step size based on the data sequence, and the average value of the data in the sliding window is recorded after each sliding.

[0009] According to a method for online fault diagnosis of a power electronic converter provided by the present invention, large-step sliding equidistant averaging processing is performed on the data sequence, comprising: A sliding window is set, and the denoised data sequence is slid with the size of the sliding window as the step size, and the average value of the data in the sliding window is recorded after each sliding.

[0010] According to a method for online fault diagnosis of a power electronic converter provided by the present invention, the initial classifier is subjected to dual-core extreme learning machine hyperparameter common factor optimization training, comprising: Obtain a sample data set with training labels and a target data set without training labels; Performing multi-step sliding average processing on the sample data set and the target data set to obtain lightweight sample features and target features; Performing initial dual-core extreme learning machine training on the initial classifier through the lightweight sample features to extract super-parameter common factors; Based on the target features, the extracted super-parameter common factors are optimized by a particle swarm algorithm, and the online classifier is generated based on the optimal super-parameter common factors.

[0011] According to a method for online fault diagnosis of a power electronic converter provided by the present invention, the initial classifier is trained with an initial dual-core extreme learning machine through the lightweight sample features to extract the hyper-parameter common factors, including: Measuring the similarity between the lightweight sample features by a kernel function, and defining a kernel matrix of an extreme learning machine; Based on the kernel matrix and the random feature mapping matrix of the lightweight sample features, a hyperparameter common factor is extracted.

[0012] According to a method for online fault diagnosis of a power electronic converter provided by the present invention, the similarity between the lightweight sample features is measured by a kernel function, including: Determine the distance between sample points of the lightweight sample feature by using a Gaussian kernel function; And / or, mapping low-dimensional spatial data in the lightweight sample features to an infinite-dimensional feature space through a polynomial kernel function to determine the similarity between the lightweight sample features.

[0013] According to a method for online fault diagnosis of a power electronic converter provided by the present invention, the extracted super-parameter common factors are optimized by a particle swarm algorithm, comprising: Initializing parameters of the particle swarm algorithm and the dual-core extreme learning machine; Determine the fitness of the particle based on a pre-set fitness function; Obtaining a local optimal solution and a global optimal solution of the particle; The speed and position of the particle are updated based on the local optimal solution and the global optimal solution of the particle until a preset termination condition is reached to obtain an optimal super-parameter common factor.

[0014] In a second aspect, the present invention further provides an online fault diagnosis device for a power electronic converter, comprising: An acquisition module, used for acquiring a data sequence of a three-phase output current output by a target converter; A processing module, used for performing multi-step sliding average processing on the data sequence to obtain a feature matrix; The diagnosis module is used to call an online classifier to perform fault diagnosis on the target converter based on the characteristic matrix and generate a fault diagnosis result; the online classifier is obtained by training a pre-built initial classifier through a dual-core extreme learning machine hyperparameter common factor optimization.

[0015] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the online fault diagnosis method for the power electronic converter as described in the first aspect above is implemented.

[0016] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for online fault diagnosis of a power electronic converter as described in the first aspect above.

[0017] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method for online fault diagnosis of a power electronic converter as described in the first aspect above.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The online fault diagnosis method for power electronic converter provided by the present invention has the advantages of high precision and high robustness by adopting the dual-core extreme learning machine hyperparameter common factor optimization for training to obtain an online classifier. In addition, since the feature matrix has been subjected to denoising and redundant data optimization processing, the data size is reduced, and the processing difficulty of the online classifier can be reduced, thereby improving the speed of fault diagnosis of the target converter, and solving the problem of low fault diagnosis efficiency existing in the existing related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is a flow chart of the online fault diagnosis method of the power electronic converter provided by the present invention; Figure 2 is a schematic diagram of performing multi-step sliding average processing in an embodiment of the present invention; Figure 3 is a schematic diagram of performing online fault diagnosis in an embodiment of the present invention; Figure 4 is a schematic diagram of the network structure of DELM in an embodiment of the present invention; Figure 5 Flow chart of the PSO algorithm for optimizing DELM hyperparameter common factors in an embodiment of the present invention; Figure 6 is a topological schematic diagram of a 4L-ANPC converter in an embodiment of the present invention; Figure 7 This is a schematic diagram of the operating state of the "F2" mark in an embodiment of the present invention; Figure 8 This is a schematic diagram of the operating state of the "F9" mark in an embodiment of the present invention; Fig. 9 Schematic diagram of the preferred super-parameter common factor in the embodiment of the present invention; Fig.10 is a schematic diagram of the results of the fault diagnosis accuracy in an embodiment of the present invention; Fig.11is a schematic diagram of a fault diagnosis confusion matrix in an embodiment of the present invention; Fig.12 It is a structural block diagram of the online fault diagnosis device for power electronic converter provided by the present invention; Fig.13 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The present invention provides a method for online fault diagnosis of a power electronic converter. Figure 1 is a flow chart of the online fault diagnosis method for power electronic converter provided by the present invention, such as Figure 1 As shown, the method comprises the following steps: Step S101, obtaining a data sequence of three-phase output current output by a target converter; Step S102, performing multi-step sliding average processing on the data sequence to obtain a feature matrix; Step S103, calling an online classifier, performing fault diagnosis on the target converter based on the feature matrix, and generating a fault diagnosis result. The online classifier is obtained by training a pre-built initial classifier through a dual-core extreme learning machine hyperparameter common factor optimization.

[0023] In this method, first, the data sequence of the three-phase output current output by the target converter is obtained. Then, the noise in the data sequence is removed by multistep sliding average processing (MSAP), the data redundancy is reduced, and the accuracy of the fault diagnosis of the target converter is improved. Then, the online classifier is called to perform fault diagnosis on the target converter. The online classifier is obtained by training the pre-built initial classifier through dual-kernel extreme learning machine-super-parametric common factor optimization (DELM-SCFO). The kernel extreme learning machine (KELM) is a simple and efficient single hidden layer feedforward neural network learning algorithm with the characteristics of simple network and extremely fast learning speed, which meets the requirements of online fault diagnosis of converters. Therefore, in this method, the online classifier obtained by training with dual-kernel extreme learning machine-super-parametric common factor optimization has the advantages of high precision and high robustness. Moreover, since the feature matrix has been processed by denoising and redundant data optimization, the data size is reduced, which can reduce the processing difficulty of the online classifier, thereby improving the speed of fault diagnosis of the target converter and solving the problem of low fault diagnosis efficiency in existing related technologies.

[0024] In some of the embodiments, without losing data features, in step S101, a multi-step sliding average process is performed on the data sequence to obtain a feature matrix, including: performing a unit step sliding average process on the data sequence to remove noise points in the data sequence; performing a large step sliding equidistant average process on the data sequence to obtain a feature matrix of the data sequence.

[0025] Specifically, the data sequence is subjected to unit step sliding average processing, including: setting a sliding window, sliding at a unit step based on the data sequence, and recording the average value of the data in the sliding window after each sliding.

[0026] Correspondingly, a large-step sliding equidistant averaging process is performed on the data sequence, including: setting a sliding window, sliding the denoised data sequence with the size of the sliding window as the step size, and recording the average value of the data in the sliding window after each sliding.

[0027] For example, Figure 2 Schematic diagram of multi-step sliding average processing in an embodiment of the present invention. Figure 2As shown in the figure, the unit step sliding average processing process is as follows: let the sliding window size be K, slide with unit step size based on the measured three-phase output current delay sequence, calculate the average value of the data in the window for each slide, and generate the denoised data measurement signal. The large step sliding equidistant average processing process is as follows: let the sliding window size be K', slide with K' step size based on the denoised three-phase output current delay sequence, and calculate the average value of the data in the window for each slide.

[0028] The proposed MSAP method overcomes the measurement noise of the 4L-ANPC converter by sliding the output current through a unit window. The large-step sliding equidistant average effectively reduces the scale of the measurement signal, improves the processing speed of the data-driven method, improves the data quality, and reduces the training difficulty of the high-precision 4L-ANPC converter fault online classifier.

[0029] Figure 3 Schematic diagram of online fault diagnosis in an embodiment of the present invention. Figure 3 As shown, in some embodiments, in order to solve the difficult problem of high-precision online fault diagnosis of the converter, the initial classifier is trained with a dual-kernel extreme learning machine for optimal hyper-parameter common factors, including: obtaining a sample data set with training labels and a target data set without training labels; performing multi-step sliding average processing on the sample data set and the target data set to obtain lightweight sample features and target features; performing initial dual-kernel extreme learning machine (DELM) training on the initial classifier through the lightweight sample features to extract hyper-parameter common factors; based on the target features, optimizing the extracted hyper-parameter common factors through a particle swarm algorithm, and generating an online classifier based on the optimally selected hyper-parameter common factors.

[0030] In this embodiment, an initial dual-kernel extreme learning machine is trained on the initial classifier through lightweight sample features to extract hyper-parameter common factors, including: measuring the similarity between lightweight sample features through a kernel function to define the kernel matrix of the extreme learning machine; and extracting hyper-parameter common factors based on a random feature mapping matrix of the kernel matrix and lightweight sample features.

[0031] Furthermore, the similarity between lightweight sample features is measured by a kernel function, including: determining the distance between sample points of the lightweight sample features by a Gaussian kernel function; and / or mapping the low-dimensional space data in the lightweight sample features to an infinite-dimensional feature space by a polynomial kernel function to determine the similarity between the lightweight sample features.

[0032] For example, for a given N sample data set {( x i , t i )}N i=1, where is the input vector, is the corresponding expected output vector. KELM measures the similarity between samples through the kernel function and defines the kernel matrix of the extreme learning machine (ELM) according to the Mercer condition, which is expressed as:

[0033] in, represents the kernel matrix, H represents the random feature map matrix, h (·) represents the feature mapping function, K (·) represents the kernel function. The output of the kernel extreme learning machine is expressed as:

[0034] in, f ( x ) represents the output of the kernel extreme learning machine, H represents the random feature map matrix, c is the regularization coefficient, β =[ β 1, β 2, …, β L ] T represents the weight matrix between the output layer and the hidden layer, T =[ t 1, t 2, …, t N ] T Represents the expected output matrix of the training sample. Kernel matrix It is only necessary to pre-select the kernel function without defining the mapping function and the number of hidden layer neurons, thus saving time for model optimization. The versatility of the kernel function directly determines the modeling capability of KELM.

[0035] The Gaussian kernel function measures data similarity by calculating the distance between sample points, which can be expressed as:

[0036] in, K 1( x i , x j ) represents the Gaussian kernel function, and g is the bandwidth parameter of the Gaussian kernel.

[0037] The polynomial kernel function enables the linear model to handle nonlinear problems by mapping low-dimensional spatial data to infinite-dimensional feature space, which can be expressed as:

[0038] in, K 2( x i , x j ) represents the polynomial kernel function, is the coefficient of the kernel function, r is a constant term. Preferably, in this embodiment, Gaussian kernel function and polynomial kernel function are used to construct dual-kernel learning rules, such as Figure 4 As shown, Figure 4 Schematic diagram of the network structure of DELM in an embodiment of the present invention, which is constructed as follows:

[0039] in, K ’ ( x i , x j ) is the kernel function of DELM, is a hyperparameter g , , r , a 1. a Common factors of 2, a 1. a 2 is the coefficient of the dual-kernel function, a 2+ a 2=1. Therefore, the intelligent optimization algorithm is used to select That is, multiple hyperparameters can be optimized at the same time. Compared with the time-consuming problem of optimizing multiple hyperparameters at the same time, the optimization of the common factors of the hyperparameters involved is fast and effective.

[0040] Particle swarm optimization (PSO) is a classic evolutionary computing technology, which has the advantages of being simple and easy to implement and not requiring many parameter adjustments. Therefore, based on the above embodiments, the PSO algorithm is used to optimize the DELM hyperparameter common factors to achieve the fusion and optimization of the polynomial kernel function and the Gaussian kernel function. The extracted hyperparameter common factors are optimized by the particle swarm algorithm, including: initializing the parameters of the particle swarm algorithm and the dual-core extreme learning machine; determining the fitness of the particles based on a pre-set fitness function; obtaining the local optimal solution and the global optimal solution of the particles; updating the speed and position of the particles based on the local optimal solution and the global optimal solution of the particles until the preset termination condition is reached to obtain the preferred hyperparameter common factors.

[0041] Specifically, the particles are mapped to the common factors of super parameters, and the fault diagnosis accuracy is used as the fitness function to carry out the optimization work. When the fault diagnosis accuracy is higher than 0.99, the iteration ends. Figure 5: is a flow chart of the PSO algorithm for optimizing DELM super-parameter common factors in an embodiment of the present invention. The optimization process is as follows Figure 5 As shown, the particle group The basic process of optimization in -dimensional space is: 1. Initialize parameters and randomly generate particles i ( i =1, …, N ) speed v i =( v i1 ,v i2 ,…, v iD ) and location x i =( x i1 ,x i2 ,…, x iD ); 2. Based on the fitness function, calculate each particle i The fitness value of F [ i ]; 3. Obtain the global optimal solution and local optimal solution of particles:

[0042]

[0043] in, p best ( i ) is a particle i The local optimal solution of g is the global optimal solution; 4. Update the speed and position of particles:

[0044]

[0045] in, r and s is a random constant, k =1,…,H is the number of iterations, c 1 and c 2 is a constant, w is the inertia constant; 5. Iterate steps 2 to 4 until the algorithm termination condition is reached.

[0046] In order to verify the effectiveness of the above method, the following actual test is carried out. The test object adopts Figure 6The 4L-ANPC converter shown, Figure 6 4L-ANPC converter according to an embodiment of the present invention. U dc is the DC side voltage, U C1 , U C2 and U C3 The DC side capacitors are C 1. C 2 and C 3 voltage. S xj ( x = { a , b , c}, j = 1, 2, …, 6) x Xiangdi j Switching devices, among which, S x1 , S x3 , S x5 Respectively S x2 , S x4 , S x6 For complementary switches, i x is the output current, L x and R x They are inductive load and linear load respectively, and the defined OCFs are shown in Table 1: Table 1 Schematic diagram of power switch tube OCFs label

[0047] In order to evaluate the diagnosis results of this method, Ac, Pr, Recall and F1 are introduced as evaluation indicators. Ac represents the percentage of correctly diagnosed samples in the total number of test samples; Pr is the ratio of samples with positive diagnosis results to actual positive samples; Recall represents the ratio of the number of actual positive samples in the positive diagnosis results to the total number of positive samples; F1 is a combination of Recall and Pr, which is used to comprehensively evaluate the diagnosis results of the proposed method. The specific expression is as follows:

[0048]

[0049]

[0050]

[0051] In the formula, TN represents the correctly diagnosed counterexample; FN represents the incorrectly diagnosed counterexample; TP represents the correctly diagnosed positive example, and FP represents the incorrectly diagnosed positive example.

[0052] Taking fault identification "F2" and "F9" as examples, the characteristic waveform of the three-phase output current is as follows: Figure 7 and Figure 8 As shown, Figure 7 is a schematic diagram of the operating state of the "F2" mark in an embodiment of the present invention, Figure 8 Schematic diagram of the running state of the "F9" mark in the embodiment of the present invention. At the same time, the MSAP method parameters K=1999, K'=50 are preset to reduce noise and lighten the three-phase output current, where the dotted line is the three-phase output current after the MSAP method is processed.

[0053] The three-phase output current processed by MSAP is divided into a training set and a test set in a ratio of 1:1. Based on the training set, the PSO method optimizes the DKLM hyperparameter common factors and optimizes the iterations such as Fig. 9 As shown, Fig. 9 is a preferred schematic diagram of the hyperparameter common factors in the embodiment of the present invention, and the preferred hyperparameter common factors =2.099.

[0054] Based on the optimized hyperparameter common factors, the 4L-ANPC converter OCFs online fault diagnosis model is established. The diagnosis results are shown in Fig.10 and Fig.11 As shown, Fig.10 is a schematic diagram of the results of the fault diagnosis accuracy in an embodiment of the present invention, Fig.11 It is a schematic diagram of the fault diagnosis confusion matrix in an embodiment of the present invention. In the figure, the higher the degree of overlap, the higher the diagnostic accuracy. It can be seen that the model misdiagnoses mainly fault categories "4", "5" and "6", and the maximum misdiagnosis rate reaches 4.63%. But overall, the online diagnosis accuracy is 99.36%, and the misdiagnosis rate is 0.64%, which still has excellent diagnostic capabilities. In particular, the 4L-ANPC converter still achieves zero false alarms when operating in a healthy state, and has excellent robustness.

[0055] In summary, the MSAP method designed in this method removes data noise and reduces data redundancy; different from the existing method of optimizing multiple hyperparameters simultaneously, the SCFO method extracts the common factors of multiple hyperparameters of DELM, and uses the particle swarm method to carry out the optimization work, thereby quickly and effectively constructing a high-precision fault classifier to improve the converter online fault diagnosis capability and accuracy. For single-switch and multi-switch OCFs, based on the four-level converter online diagnosis platform built in the laboratory, the experimental results show that the proposed method can effectively diagnose four-level converter OCFs, with an accuracy rate higher than 99.35% and a maximum diagnostic error lower than 5%. It has the advantages of simple operation, good real-time performance, and strong generalization ability.

[0056] The present invention also provides an online fault diagnosis device for a power electronic converter. The online fault diagnosis device for a power electronic converter provided by the present invention is described below. The online fault diagnosis device for a power electronic converter described below and the online fault diagnosis method for a power electronic converter described above can be referred to each other. Fig.12 is a structural block diagram of the online fault diagnosis device for power electronic converter provided by the present invention, such as Fig.12 As shown, the device comprises: An acquisition module 1201 is used to acquire a data sequence of a three-phase output current output by a target converter; A processing module 1202 is used to perform multi-step sliding average processing on the data sequence to obtain a feature matrix; The diagnosis module 1203 is used to call the online classifier, perform fault diagnosis on the target converter based on the feature matrix, and generate a fault diagnosis result; the online classifier is obtained by training the pre-built initial classifier through the dual-core extreme learning machine hyperparameter common factor optimization.

[0057] When the device is in use, first, the acquisition module 1201 acquires the data sequence of the three-phase output current output by the target converter. Then, the processing module 1202 performs multi-step sliding average processing to remove the noise in the data sequence, reduce data redundancy, and improve the accuracy of fault diagnosis of the target converter. Then, the diagnosis module 1203 calls the online classifier to perform fault diagnosis on the target converter. The online classifier is obtained by training the pre-built initial classifier through the super-parameter common factor optimization of the dual-core extreme learning machine. The nuclear extreme learning machine is a simple and efficient single-hidden layer feedforward neural network learning algorithm, which has the characteristics of simple network and extremely fast learning speed, and meets the requirements of online fault diagnosis of the converter. Therefore, in the above process, the online classifier obtained by training with the super-parameter common factor optimization of the dual-core extreme learning machine has the advantages of high precision and high robustness. In addition, since the feature matrix has been subjected to denoising and redundant data optimization processing, the data size is reduced, and the processing difficulty of the online classifier can be reduced, thereby improving the speed of fault diagnosis of the target converter, and solving the problem of low fault diagnosis efficiency in the existing related technologies.

[0058] Fig.13 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.13 As shown, the electronic device may include: a processor 1301, a communications interface 1302, a memory 1303 and a communication bus 1304, wherein the processor 1301, the communications interface 1302, and the memory 1303 communicate with each other through the communication bus 1304. The processor 1301 may call the logic instructions in the memory 1303 to execute the online fault diagnosis method of the power electronic converter, the method comprising: Acquire a data sequence of three-phase output current output by the target converter; Perform multi-step sliding average processing on the data sequence to obtain the feature matrix; The online classifier is called to perform fault diagnosis on the target converter based on the feature matrix to generate fault diagnosis results; the online classifier is obtained by training the pre-built initial classifier through the dual-core extreme learning machine hyperparameter common factor optimization.

[0059] In addition, the logic instructions in the above-mentioned memory 1303 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0060] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the power electronic converter online fault diagnosis method provided by the above methods, the method includes: Acquire a data sequence of three-phase output current output by the target converter; Perform multi-step sliding average processing on the data sequence to obtain the feature matrix; The online classifier is called to perform fault diagnosis on the target converter based on the feature matrix to generate fault diagnosis results; the online classifier is obtained by training the pre-built initial classifier through the dual-core extreme learning machine hyperparameter common factor optimization.

[0061] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for online fault diagnosis of a power electronic converter provided by the above methods is implemented, and the method comprises: Acquire a data sequence of three-phase output current output by the target converter; Perform multi-step sliding average processing on the data sequence to obtain the feature matrix; The online classifier is called to perform fault diagnosis on the target converter based on the feature matrix to generate fault diagnosis results; the online classifier is obtained by training the pre-built initial classifier through the dual-core extreme learning machine hyperparameter common factor optimization.

[0062] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0063] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for online fault diagnosis of a power electronic converter, characterized in that: include: Acquire a data sequence of three-phase output current output by the target converter; Performing multi-step sliding average processing on the data sequence to obtain a feature matrix; An online classifier is called to perform fault diagnosis on the target converter based on the characteristic matrix to generate a fault diagnosis result; the online classifier is obtained by training a pre-built initial classifier through a dual-core extreme learning machine hyperparameter common factor optimization.

2. The method for online fault diagnosis of power electronic converter according to claim 1, characterized in that: The data sequence is subjected to multi-step sliding average processing to obtain a feature matrix, including: Performing unit step sliding average processing on the data sequence to remove noise points in the data sequence; The data sequence is subjected to large-step sliding equidistant averaging processing to obtain a characteristic matrix of the data sequence.

3. The method for online fault diagnosis of power electronic converter according to claim 2, characterized in that: The data sequence is subjected to unit step sliding average processing, including: A sliding window is set, which slides with a unit step size based on the data sequence, and the average value of the data in the sliding window is recorded after each sliding.

4. The method for online fault diagnosis of power electronic converter according to claim 2, characterized in that: The data sequence is subjected to large-step sliding equidistant averaging processing, including: A sliding window is set, and the denoised data sequence is slid with the size of the sliding window as the step size, and the average value of the data in the sliding window is recorded after each sliding.

5. The method for online fault diagnosis of power electronic converter according to claim 1, characterized in that: The initial classifier is subjected to dual-core extreme learning machine hyperparameter common factor optimization training, including: Obtain a sample data set with training labels and a target data set without training labels; Performing multi-step sliding average processing on the sample data set and the target data set to obtain lightweight sample features and target features; Performing initial dual-core extreme learning machine training on the initial classifier through the lightweight sample features to extract super-parameter common factors; Based on the target features, the extracted super-parameter common factors are optimized by a particle swarm algorithm, and the online classifier is generated based on the optimal super-parameter common factors.

6. The method for online fault diagnosis of power electronic converter according to claim 5, characterized in that: The initial classifier is trained with an initial dual-core extreme learning machine through the lightweight sample features to extract super-parameter common factors, including: Measuring the similarity between the lightweight sample features by a kernel function, and defining a kernel matrix of an extreme learning machine; Based on the kernel matrix and the random feature mapping matrix of the lightweight sample features, a hyperparameter common factor is extracted.

7. The method for online fault diagnosis of power electronic converter according to claim 6, characterized in that: The similarity between the lightweight sample features is measured by a kernel function, including: Determine the distance between sample points of the lightweight sample feature by using a Gaussian kernel function; And / or, mapping low-dimensional spatial data in the lightweight sample features to an infinite-dimensional feature space through a polynomial kernel function to determine the similarity between the lightweight sample features.

8. The method for online fault diagnosis of power electronic converter according to claim 5, characterized in that: The extracted super-parameter common factors are optimized by using a particle swarm algorithm, including: Initializing parameters of the particle swarm algorithm and the dual-core extreme learning machine; Determine the fitness of the particle based on a pre-set fitness function; Obtaining a local optimal solution and a global optimal solution of the particle; The speed and position of the particle are updated based on the local optimal solution and the global optimal solution of the particle until a preset termination condition is reached to obtain an optimal super-parameter common factor.

9. An online fault diagnosis device for a power electronic converter, characterized in that: include: An acquisition module, used for acquiring a data sequence of a three-phase output current output by a target converter; A processing module, used for performing multi-step sliding average processing on the data sequence to obtain a feature matrix; The diagnosis module is used to call an online classifier to perform fault diagnosis on the target converter based on the characteristic matrix and generate a fault diagnosis result; the online classifier is obtained by training a pre-built initial classifier through a dual-core extreme learning machine hyperparameter common factor optimization.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the online fault diagnosis method for the power electronic converter according to any one of claims 1 to 8 is implemented.

Images